A
The graphical user interface in Cadence Virtuoso used to configure simulation models, analyses (transient, DC, AC), design variables, corners, and output expressions.
In this project: ADE L session states were scripted into batch OCEAN scripts (`run_schematic_tb_param.ocn`) to execute automated Spectre runs on `divider_lib/tb_divider` without launching the GUI.
The Antigravity CLI and agent runtime developed by Google DeepMind for advanced agentic pair-programming and multi-agent coordination.
In this project: Used to coordinate high-level architecture, execute prose audits, synthesize technical documentation, and verify simulation consistency against source evidence.
Cadence standard primitive component library containing ideal electrical elements such as independent sources (vpulse, vdc), ground symbols (gnd!), and ideal passive resistors (res) and capacitors (cap).
In this project: Used in `divider_lib` schematics for supply sources (`vdc`), differential clock stimuli (`vpulse`), and ideal load resistors (`res`). The ideal 2-terminal `res` later created an LVS mismatch against the 3-terminal PDK resistor `g45rnsnp`.
Plasma-induced gate oxide damage occurring during reactive ion etching when long, unconnected metal interconnect lines accumulate charge that discharges through thin transistor gate oxides.
In this project: Monitored during Assura DRC checks; in our placement layout, transistor gates are kept at unit lengths and connected to landing pads to prevent oxide breakdown during future routing passes.
The ratio of channel width to channel length in a MOSFET, governing current drive capacity ($I_{ds} \propto W/L$), transconductance ($g_m$), and parasitic gate capacitance.
In this project: Unit NMOS devices (M1–M6) use $W/L = 240\text{ nm} / 45\text{ nm} = 5.33$, closely matching Kuo et al.'s reference ratio of $5.00$ ($900\text{ nm} / 180\text{ nm}$). Tail device M7 uses $480\text{ nm} / 90\text{ nm} = 5.33$.
Cadence physical verification suite performing Design Rule Checking (DRC), Layout Versus Schematic (LVS), and Parasitic Extraction (RCX) on IC layouts.
In this project: Assura 4.1 was run in batch mode inside the CentOS VM via `run_verify.sh` to check `divider_lib/DLATCH` and `divider_core` layouts against `assuraDRC.rul` and `extract.rul`.
B
A frozen, verified state of repository files, netlists, reports, and VM images that serves as an authoritative reference point for future work.
In this project: Recorded in STABLE.md at git tag `v1.0-stable` (commit 336d17b) and VM snapshot `stable-v1`, preserving verified 45 nm metrics for Designs D and B so v2 research proceeds independently on `main`.
A steady DC control potential applied to the gate of an MOS transistor to set its quiescent gate-to-source bias ($V_{gs}$) and determine its operating current.
In this project: Nominal tail bias $v_b = 1.0\text{ V}$ is applied to M7's gate. Swept between $0.8\text{ V}$ and $1.0\text{ V}$ during initial sizing exploration; $1.0\text{ V}$ was selected to maximize pull-down drive headroom.
The shift in a MOSFET's threshold voltage ($V_{th}$) caused by a non-zero voltage difference between source and substrate ($V_{sb} > 0$), widening the depletion region.
In this project: In the stacked NMOS sampling network (M1 above M5 above M7), sources of M1 and M5 sit above ground ($V_s > 0$), raising their effective $V_{th}$ and increasing stack on-resistance.
C
The industry-standard electronic design automation (EDA) software platform by Cadence Design Systems for schematic entry, layout editing, and mixed-signal IC design.
In this project: Virtuoso IC615 (6.1.5.500.15) was executed inside the CentOS VM to generate schematics, symbols, Layout XL placements, and routing views for `divider_lib`.
A headless 32-bit CentOS 6.5 VirtualBox guest machine used to host the Cadence IC615, Spectre 12.1, and Assura 4.1 software suite.
In this project: Driven from the Linux host through an automated pipeline utilizing headless serial port communication (`vmx.sh`, `serial.txt`) and dynamic ISO staging (`xorriso`).
Cadence proprietary data structure that defines component parameters, simulation models, terminal labels, and callback procedures for Virtuoso library cells.
In this project: Interacted with via SKILL (`cdfGetInstCDF`) in `build_schematic_v2.il` to set device width (`w`), length (`l`), multiplier (`m`), and finger counts before schematic netlisting.
The shortening of the effective inverted channel length in a MOSFET as drain-to-source voltage ($V_{ds}$) increases beyond pinch-off, causing a finite output conductance $g_{ds}$.
In this project: Accounted for by the BSIM4 model in GPDK045; tail transistor M7 uses a double-length channel ($L = 90\text{ nm}$) to mitigate channel length modulation and improve output resistance.
The effective electrical resistance between drain and source terminals of an MOS transistor operating in the deep linear (triode) conduction region: $R_{on} \approx [\mu C_{ox} (W/L) (V_{gs} - V_{th})]^{-1}$.
In this project: Measured via DC sweep in GPDK045 for unit NMOS ($240\text{ nm} / 45\text{ nm}$) at $V_{gs} = 1.0\text{ V}, V_{ds} = 10\text{--}100\text{ mV}$ to be $R_{on} = 2.3\text{ k}\Omega$ ($2.28\text{--}2.73\text{ k}\Omega$).
The Cadence automated grid-based and shape-based interconnect router used to execute global, track, and detail routing passes on physical layouts.
In this project: Invoked in batch mode via `cadence/skill/auto/route_dlatch.il` on `DLATCH/layout_lxl`, routing 136 of 137 connections (99.27% completion) with 1 unrouted net (`xs`).
The primary control and console logging window of Cadence Virtuoso where user commands, SKILL expressions, load logs, and tool error messages are output.
In this project: Captured in `virtuoso_ciw_gpdk045_loaded.png` to confirm proper initialization of `gpdk045_v_6_0` and automated execution of batch SKILL scripts without interactive intervention.
Complementary differential clock signals where CKb is 180° out of phase with CK, providing non-overlapping control for sampling and regeneration phases.
In this project: Input differential clocks run at 1.4 GHz (Design D) and 6.0 GHz (Design B). Inverted clock connections (CKb to sampling, CK to hold) are applied to latches 2 and 3 in the ring.
An institutional ledger recording every numerical assertion and benchmark in a manuscript draft, linking each claim to an evidentiary simulation file and audit verdict.
In this project: Introduced in v2/CLAIMS_AUDIT.md to audit manuscript claims; enforces the rule that no metric enters the paper without an evidence path and rerunnable execution script.
Integrated circuit technology combining both n-type (NMOS) and p-type (PMOS) field-effect transistors on a single silicon substrate to achieve high noise immunity and low static power.
In this project: The core ring divider employs an NMOS current-steering topology with resistive loads to maximize switching speed and eliminate PMOS gate-loading capacitance.
A concise peer-reviewed paper documenting where a published circuit design equation breaks down under advanced process nodes, providing a revised physical model and measured data.
In this project: Designated in v2/PLAN.md as the fallback paper strategy: documenting that Kuo's R = Ron/2 rule fails at unit size in 45 nm, and demonstrating the stack rule R1 >= 1.1 * 3Ron/m with measured Designs D and B.
The longest sequence of dependent tasks that dictates the minimum calendar duration required to complete an engineering project milestone.
In this project: Outlined in v2/SCHEDULE.md as W1 (Decisions) -> W3 (Re-baseline) -> W5 (E1 Sweep) -> W7 (Analysis) -> W10 (Manuscript) -> W11 (Audits) -> W12 (Package), spanning 4 weeks with serialized VM runs.
Two active transistors where the drain of each is cross-connected to the gate of the other, forming a regenerative positive feedback loop with high differential negative resistance.
In this project: Transistors M3 and M4 in each DLATCH form the cross-coupled regeneration pair, holding and regenerating the sampled state when clock switch M6 is enabled by CKb.
D
The high-speed sized frequency divider variant designed to maximize operating frequency and achieve benchmark phase accuracy near its speed limit.
In this project: Configured with device multiplier $m = 4$, load resistors $R_1 = R_2 = 2\text{ k}\Omega$, achieving 3000.0 MHz output, $0.493\text{ mW}$ power, $27.94\text{ ps}$ delay, $13.77\text{ fJ}$ PDP, and $3.78^\circ$ phase error.
The nominal sized frequency divider variant designed for operation at the reference paper's input clock frequency of 1.4 GHz.
In this project: Configured with multiplier $m = 4$, load resistors $R_1 = R_2 = 3\text{ k}\Omega$, achieving 700.0 MHz output, $0.541\text{ mW}$ power, $25.99\text{ ps}$ delay, $14.07\text{ fJ}$ PDP, and 41/45 PVT corner yield.
Cadence executable that converts a Design Framework II (DFII) schematic view into a Volume Layout Database (VLDB) netlist file required by Assura LVS.
In this project: Called automatically within `cadence/skill/verify/run_verify.sh` using template `netlist.vlr.in` to translate `divider_lib/DLATCH` schematic into a format compatible with Assura.
A symmetrical two-transistor circuit topology that amplifies differential input voltage signals while rejecting common-mode noise and power supply fluctuations.
In this project: Transistors M1 and M2 in DLATCH form the input sampling differential pair, steering tail current between differential nodes Q and Qb based on inputs D and Db.
Geometric layout parameters specifying source/drain junction area (`as`, `ad`) and perimeter (`ps`, `pd`), directly determining junction parasitic capacitances.
In this project: Automatically calculated by Virtuoso PCell callbacks (`as=16.8f, ad=16.8f, ps=520n, pd=520n`) and injected into the netlist during schematic-driven simulations.
Parameters representing the number of diffusion squares in series with drain (`nrd`) and source (`nrs`), determining extrinsic source/drain parasitic sheet resistance.
In this project: Extracted as `nrd=1.16667, nrs=1.16667` by the Virtuoso netlister; their addition to the standalone netlist contributed to explaining the $0.541\text{ mW} \to 0.455\text{ mW}$ power gap.
A frequency division function where the fundamental frequency of the output signal is exactly half that of the input clock ($f_{out} = f_{in} / 2$).
In this project: Produces 700 MHz output from 1.4 GHz input (Design D) and 3000 MHz output from 6.0 GHz input (Design B) across all eight differential output phases.
Differential D-type latch stage comprising a differential sampling pair, a regenerative cross-coupled pair, clock gating switches, resistive loads, and a tail current device.
In this project: The fundamental building block of the 8-phase divider; contains 7 NMOS transistors (M1–M7) and 2 load resistors (R1, R2). Four instances (XD1–XD4) form the closed ring.
Physical verification process confirming that a layout geometry complies with all semiconductor foundry manufacturing constraints (widths, spacings, overlaps, enclosures).
In this project: Executed using Assura 4.1 against `assuraDRC.rul`; DLATCH placement layout produced 21 real errors across 3 rules, while auto-routed layout produced 134 errors across 17 rules.
The ratio of high-logic duration to total waveform period ($T_{high} / T_{period} \times 100\%$).
In this project: Outputs exhibit balanced $50.0\% \pm 0.5\%$ duty cycle at steady state due to matched rise and fall delays ($t_{pdr} \approx t_{pdf} = 4RC$) and symmetrical differential topology.
E
An analytical RC network delay approximation model that estimates signal propagation delay as the sum of resistances multiplied by the downstream downstream capacitances.
In this project: Applied to derive latch propagation delay: $t_{pdf} = (8C)(R/2) = 4RC$ and $t_{pdr} = (8C)(R_1) = 4RC$, demonstrating symmetric charging and discharging dynamics.
F
The critical project milestone where synthetic agent-generated baseline data and an empty 178-byte GDSII layout were identified, rejected, and replaced with verified Cadence simulation.
In this project: In commit `a805b27`, all hardcoded scripts and fabricated results were eliminated, establishing real Spectre MMSIM121 and Virtuoso IC615 verification pipelines in GPDK045.
The number of standard logic gate inputs driven by a single output node, determining effective capacitive loading and signal transition slew rates.
In this project: Each DLATCH output in the ring drives exactly one subsequent differential input pair ($FO = 1$) plus an assumed $10\text{ fF}$ interconnect and probing testbench capacitance.
A unit of electrical energy equal to $10^{-15}$ Joules ($1\text{ fJ} = 1\text{ mW} \times 1\text{ ps}$), commonly used to quantify power-delay product in nanoscale VLSI circuits.
In this project: Standard unit for PDP across all report tables: $14.067\text{ fJ}$ for Design D, $13.768\text{ fJ}$ for Design B, and $8.557\text{ fJ}$ for the Unsized Baseline.
A composite performance index that combines disparate circuit metrics (operating frequency, power consumption, area, phase error) into a single scalar value for cross-technology comparison.
In this project: Used in published divider benchmark tables (Kuo et al., Table I); v2 requirements require primary source retrieval for all benchmark entries rather than copying unverified table rows.
A composite performance metric in frequency divider literature balancing operating frequency, power dissipation, and phase accuracy: $\text{FoM}_\phi = 10 \log_{10}[f_{in} / (P \cdot \Delta\phi_{max})]$.
In this project: Calculated to compare 45 nm GPDK045 implementation efficiency against published 180 nm benchmark data from Kuo et al.
A POSIX file advisory locking mechanism used on the host operating system to serialize access to shared hardware resources.
In this project: Used via `flock vm.lock` across all host automation scripts (`vmx.sh`, `skill_run.sh`, `run_verify.sh`) to prevent concurrent execution conflicts on the VirtualBox guest.
Dividing a wide transistor channel into multiple parallel narrower fingers sharing common source and drain diffusions to reduce gate resistance and diffusion capacitance.
In this project: Used in Layout XL generation where multiplier $m = 4$ devices were instantiated as 4 parallel fingers, resulting in 28 active MOS channels in view `layout_lxl`.
G
The standard-threshold, 1.0 V nominal thin-oxide NMOS device provided in Cadence GPDK045, represented in Virtuoso schematics by cell `nmos1v`.
In this project: All active transistors (M1–M7) in `divider_lib` are instantiated using this model section with $W = 240\text{ nm}, L = 45\text{ nm}$ ($L = 90\text{ nm}$ for M7).
The non-silicided n-plus polysilicon resistor cell in Cadence GPDK045, featuring 3 electrical terminals (plus, minus, and substrate bulk tie).
In this project: Used in the physical layout placement of DLATCH for load resistors R1 and R2; requires a substrate connection that caused device mismatches against 2-terminal ideal schematic resistors.
The small-signal output conductance of a transistor ($g_{ds} = \partial I_{ds} / \partial V_{ds}$), representing the inverse of small-signal output channel resistance $r_o$.
In this project: Higher $g_{ds}$ in 45 nm short-channel devices lowers the intrinsic gain $g_m / g_{ds}$, necessitating higher load resistance values ($R_1$) to satisfy regeneration loop gain.
Virtuoso Layout Suite XL command that automatically translates schematic instances and connectivity bindings into placed layout PCells and boundary pins.
In this project: Executed in batch via `cadence/skill/auto/lxl_flow.il` on `DLATCH` schematic, generating view `layout_lxl` containing 28 NMOS transistors and 9 I/O pins.
The ratio of change in drain current to the change in gate-to-source voltage ($g_m = \partial I_{ds} / \partial V_{gs}$), measuring device amplification strength.
In this project: Regeneration requires loop gain $g_m R_1 > 1$ in the cross-coupled pair M3–M4. Sizing with multiplier $m = 4$ increases total $g_m$, enabling division with smaller load resistance.
A mandatory review milestone between project stages where strict technical criteria must be fulfilled before committing time and VM resources to subsequent tasks.
In this project: Structured across stages S0 to S5 in v2/PLAN.md; for example, Stage S1 exit requires successful nominal rerun of Designs D and B before launching Stage S2 enhancement sweeps.
Cadence Generic Process Design Kit for 45 nm CMOS technology, providing BSIM4 SPICE models, Virtuoso PCells, techfiles, and Assura verification rule decks.
In this project: The primary process node for this project (`gpdk045_v_6_0`), evaluating thin-oxide 1.0 V standard devices under nominal and PVT extreme corners.
The common reference potential node (0 V) in the circuit.
In this project: Virtuoso schematics require global ground net `gnd!` using the `analogLib/gnd` symbol. Standalone Spectre decks reference node `0`.
H
Lightweight, rapid AI worker agents deployed in background sessions to execute repetitive commands inside the Cadence VM.
In this project: Dispatched to automate command replays, file shipping, log parsing, and DRC error extraction under the coordination of the lead model.
I
Signal inversion logic function where logic high maps to low and vice versa.
In this project: In the differential ring divider, inversion is implemented without active inverters simply by cross-coupling complementary differential pairs ($Q_4 \to D_{1b}, Q_{4b} \to D_1$).
J
Undesired short-term variations and phase fluctuations of the significant clock transition edges from their ideal positions in time.
In this project: Minimized by symmetric differential switching and equalized rise/fall times; simulated waveforms exhibit sharp transitions with jitter $< 1.2\text{ ps}$.
L
A destructive low-impedance short-circuit path formed between power and ground rails by parasitic bipolar transistors (p-n-p-n SCR structures) in bulk CMOS.
In this project: Governed by rule `LATCHUP.2` requiring substrate taps within maximum allowable distances from diffusion; router attempt triggered 56 `LATCHUP.2` violations due to tap spacing.
Assura DRC rule in GPDK045 mandating that NMOS source/drain diffusions must reside within a specified proximity to a p-substrate tie tap.
In this project: The single largest category of DRC violations (56 errors) produced during the automatic router attempt, requiring dedicated guard-ring and tap placement.
Physical verification stage that compares extracted circuit connectivity from layout against the original schematic netlist to verify 1-to-1 equivalence.
In this project: Assura LVS run via `run_verify.sh`; reported device unbound errors due to 2-terminal schematic `analogLib/res` versus 3-terminal layout `g45rnsnp` resistors, plus open net `xs`.
Connectivity-aware layout environment in Cadence Virtuoso that maintains real-time binding with the schematic, enabling guided placement, flying wires, and DRC checking.
In this project: Utilized in batch mode via `lxGenFromSource` (`cadence/skill/auto/lxl_flow.il`) to generate bound layout views with proper device terminals and nets.
The total lumped capacitive load present at an output node, consisting of gate oxide capacitances, junction diffusion capacitances, and interconnect parasitics.
In this project: Fixed at $10\text{ fF}$ per output terminal across all nominal and PVT characterization testbenches to model downstream loading and pad capacitance.
Passive pull-up resistive elements connected between the positive supply rail (VDD) and the differential output nodes (Q, Qb).
In this project: Selected as $8\text{ k}\Omega$ for Unsized Baseline, $3\text{ k}\Omega$ for Design D, and $2\text{ k}\Omega$ for Design B; direct implementation of paper value ($1.15\text{ k}\Omega$) fails to divide.
M
Differential NMOS transistor pair responsible for sensing input signals D and Db during the clock sampling phase.
In this project: Configured with $W = 240\text{ nm}, L = 45\text{ nm}$, multiplier $m = 4$ in Design D. Source nodes connect to clock switch transistor M5 at intermediate node `xs`.
Cross-coupled NMOS transistor pair providing positive regenerative feedback to latch and store logic states during hold phase.
In this project: Configured identically to sampling devices with $W = 240\text{ nm}, L = 45\text{ nm}$, multiplier $m = 4$ in Design D. Sources connect to switch M6 at node `xh`.
Clock-controlled switch transistor activated when CK is high to connect the sampling pair M1–M2 to the tail node.
In this project: Gate connected to clock signal CK; drain connects to sampling pair common source `xs`, source connects to tail node `tail`.
Clock-controlled switch transistor activated when complementary clock CKb is high to enable regenerative latching.
In this project: Gate connected to complementary clock CKb; drain connects to cross-coupled pair common source `xh`, source connects to tail node `tail`.
NMOS transistor connected between common tail node and ground, biased at gate voltage vb to establish total bias current.
In this project: Sized with channel length $L = 90\text{ nm}$ and width $W = 480\text{ nm}$. Operates in triode region ($V_{ds} = 8\text{--}67\text{ mV} < V_{dsat} = 358\text{ mV}$) across all designs.
Planar conductive aluminum/copper interconnect layers separated by dielectric insulating oxide in a CMOS process stack.
In this project: GPDK045 features 11 metal layers. Metal1 is used for internal cell pins and routing; Metal2 is used for inter-latch ring connections.
Assura DRC rule specifying that any polygon drawn on Metal1 must have a minimum planar surface area of at least $0.02\text{ }\mu\text{m}^2$.
In this project: Produced 14 real errors in the DLATCH placement layout where small square contact pin shapes fell below the minimum area constraint.
The minimum load resistance required to establish a regenerative loop gain greater than unity ($g_m R_L > 1$) in a differential latch.
In this project: Derived as $R_{crit} = 1.1 \times (3R_{on} / m) = 7.59\text{ k}\Omega$ for $m=1$, and $1.90\text{ k}\Omega$ for $m=4$. Explains the failure of Kuo et al.'s $1.15\text{ k}\Omega$ rule.
The drift velocity acquired by charge carriers (electrons in NMOS, holes in PMOS) per unit applied electric field: $v_d = \mu E$.
In this project: Higher electron mobility ($\mu_n \approx 3\times \mu_p$) motivated an all-NMOS pull-down switching network with resistive loads for high-frequency operation.
Named configuration blocks within SPICE model libraries defining statistical process corner variations for NMOS and PMOS transistors.
In this project: Specified in Spectre runs via `section=tt` (typical-typical), `section=ff` (fast-fast), `section=ss` (slow-slow), `section=fs` (fast-slow), and `section=sf` (slow-fast).
A statistical circuit simulation method evaluating parametric yield by randomly varying device model parameters according to foundry-measured statistical and mismatch distributions.
In this project: Subject to requirement SR-06: permitted in v2 only after inspecting the GPDK045 Spectre model files to verify that mismatch sections exist, preventing ungrounded statistical yield claims.
Metal-Oxide-Semiconductor Field-Effect Transistor, a voltage-controlled switching and amplifying device utilizing an insulated gate.
In this project: The fundamental semiconductor device used throughout the frequency divider core in the Cadence GPDK045 45 nm technology kit.
An electrical clock generation architecture that produces multiple periodic output signals possessing equal frequency but systematically staggered phase angles.
In this project: The 4-latch divider produces eight distinct phases: $\phi_0 = 0^\circ, \phi_1 = 45^\circ, \phi_2 = 90^\circ, \phi_3 = 135^\circ, \phi_4 = 180^\circ, \phi_5 = 225^\circ, \phi_6 = 270^\circ, \phi_7 = 315^\circ$.
A parameter in Cadence schematic and SPICE netlists that instantiates multiple identical parallel copies of a transistor to scale total width.
In this project: Set to $m = 1$ in the unsized baseline, and $m = 4$ in Design D and Design B, effectively quadrupling transistor current drive and reducing stack resistance.
N
A text-based description of circuit connectivity, device models, and parameter instances used by simulation and verification engines.
In this project: Generated via Cadence Virtuoso netlister (`dfII`) to produce `results/gpdk045/virtuoso_netlist_tb_divider.scs`, diffed against standalone Spectre netlist.
N-channel Metal-Oxide-Semiconductor field-effect transistor conducting electrons through an n-type inverted channel formed in a p-type substrate.
In this project: Used exclusively for all 7 active devices in each DLATCH cell (`g45n1svt`), operating at nominal 1.0 V supply voltage.
O
Cadence specialized SKILL-based command-line scripting language designed to automate simulation setup, batch execution, and waveform post-processing.
In this project: Utilized in `cadence/ocean/run_schematic_tb_param.ocn` to automate schematic-driven simulation of `tb_divider` at 1.4 GHz directly in batch mode.
The peak-to-peak voltage differential between the minimum and maximum steady-state logic output levels ($V_{swing} = V_{max} - V_{min}$).
In this project: Measured as $0.4344\text{ V}$ ($0.583\text{--}1.017\text{ V}$) for Design D, $0.1349\text{ V}$ ($0.818\text{--}0.953\text{ V}$) for Design B, and $0.0153\text{ V}$ for paper sizing (non-dividing).
P
Direct port of Kuo et al.'s sizing rule ($m=1, R_1 = R_2 = R_{on}/2 = 1.15\text{ k}\Omega$) into GPDK045 at 1.4 GHz, which fails to divide.
In this project: Simulation confirmed output oscillates at 2799.8 MHz ($2.0\times f_{in}$) with only $15.3\text{ mV}$ swing, failing the division criteria due to sub-threshold loop gain.
Unwanted intrinsic capacitances formed by gate-to-channel overlap ($C_{gs}, C_{gd}$), channel-to-bulk ($C_{gb}$), and reverse-biased source/drain p-n junctions ($C_{sb}, C_{db}$).
In this project: Limits the maximum operating frequency of Design B to 6.0 GHz; single-edge delay of $66.1\text{ ps}$ is $3.3\times$ the intrinsic parasitic-free RC bound of $20\text{ ps}$.
A switching circuit arrangement using MOS transistors to pass or block bidirectional analog or digital signals under clock control.
In this project: Switch transistors M5 and M6 act as clock-controlled pass switches directing current between sampling/regeneration pairs and the tail current source.
A programmable Virtuoso layout or schematic generator whose geometric dimensions and terminal structures update dynamically based on CDF parameters.
In this project: All transistor placements in `divider_lib` utilize GPDK045 `nmos1v` PCells that automatically adjust diffusion shapes and contacts when $W, L, m$ are altered.
A comprehensive package of device models, technology files, PCells, and verification rule decks provided by a foundry for designing ICs in a specific fabrication process.
In this project: Cadence GPDK045 version 6.0 (`gpdk045_v_6_0`) installed in guest `gpdk045_v_6_0`.
The maximum absolute angular deviation of an actual output phase step from the ideal theoretical spacing of 45.0 degrees: $\Delta\phi_{max} = \max |\Delta\phi_i - 45^\circ|$.
In this project: Design D at 1.4 GHz exhibits $38.46^\circ$ error ($83.46^\circ$ and $6.54^\circ$ steps); Design B at 6.0 GHz achieves $3.78^\circ$ error ($48.78^\circ$ and $41.22^\circ$), matching paper benchmark.
The angular difference between consecutive output phases in an 8-phase frequency divider: $\alpha = 180 \times t_{pair} \times f_{in}$ degrees.
In this project: Calculated from inter-pair edge delay $t_{pair}$; yields alternating steps of $\alpha$ and $(90^\circ - \alpha)$ across the eight output terminals.
The geometrical polygon representation of semiconductor layers, diffusions, polysilicon gates, metal interconnects, and vias used to manufacture integrated circuits.
In this project: Created for DLATCH and divider_core using SKILL scripts; contains placed PCells and pins in `layout`, and auto-routed tracks in `layout_routed`.
An IC layout stage containing positioned transistor cells, passives, and I/O boundary pins, but lacking detailed interconnect routing between terminals.
In this project: The primary verified layout milestone in this report; contains 7 NMOS PCells and 2 poly resistors in DLATCH, and a 2x2 array in divider_core with 21 DRC errors.
P-channel Metal-Oxide-Semiconductor field-effect transistor conducting positive charge carriers (holes) through an inverted p-type channel in an n-well.
In this project: Excluded from the high-speed divider core pull-down networks to avoid carrier mobility penalties and parasitic gate loading, but available in GPDK045 as `pmos1v`.
Total electrical energy consumed per unit time by the circuit, calculated as average supply current multiplied by nominal supply voltage: $P = V_{DD} \times I_{avg}$.
In this project: Measured across steady-state clock cycles: $0.1375\text{ mW}$ for Baseline U8k, $0.5413\text{ mW}$ for Design D standalone ($0.4555\text{ mW}$ schematic-driven), and $0.4929\text{ mW}$ for Design B.
Energy consumed per switching event, calculated as the product of total power dissipation and average propagation delay: $\text{PDP} = P_{avg} \times t_{pd}$.
In this project: Computed in femtojoules (fJ): $8.557\text{ fJ}$ for Baseline U8k, $14.067\text{ fJ}$ for Design D ($12.142\text{ fJ}$ schematic-driven), and $13.768\text{ fJ}$ for Design B.
A frozen, clean virtual machine disk state stored prior to batch simulation and routing experiments, allowing instant rollback in case of disk exhaustion or environment corruption.
In this project: Maintained on the host machine to ensure reproducible Cadence Virtuoso and Spectre execution without guest image drift.
Statistical fabrication process variations representing extreme speed boundaries for NMOS and PMOS devices: typical (tt), fast-fast (ff), slow-slow (ss), fast-slow (fs), slow-fast (sf).
In this project: All 5 process corners were evaluated across temperatures and voltages in 45-run PVT sweeps; Design D passed across tt, ff, ss, fs, and sf corners.
The elapsed time between an input clock transition (50% level) and the corresponding transition at an output terminal.
In this project: Defined as the multi-phase average delay from CK rising edge to the next transition across any output: $25.99\text{ ps}$ for Design D, $27.94\text{ ps}$ for Design B, and $62.21\text{ ps}$ for Baseline.
Cadence binary simulation database format used by Spectre to store transient, DC, and AC simulation waveforms and operating point parameters.
In this project: Output generated in guest run directories (`run.raw`); converted to ASCII format or read via OCEAN expressions to extract delay, power, and nodal voltages.
Design characterization assessing circuit functionality and performance across variations in semiconductor manufacturing, supply rail, and ambient temperature.
In this project: Tested across a 45-corner matrix (5 process corners $\times$ 3 temperatures [$-40, 27, 125^\circ\text{C}$] $\times$ 3 supplies [$0.9, 1.0, 1.1\text{ V}$]). Design D achieved 41/45 pass rate.
Q
Cadence next-generation parasitic extraction tool replacing older Assura RCX, extracting accurate 3D field-solver capacitances and resistances.
In this project: Unavailable in our local GPDK045 copy because the required `qrc/` setup directory was absent, preventing post-layout extraction.
R
Text-formatted SPICE simulation output format containing structured simulation vectors, time steps, and nodal waveforms.
In this project: Generated with `spectre -format nutascii -raw` inside the VM, enabling Python script `measure_guest.py` to parse waveforms without requiring Cadence GUI licenses.
The active latching process in a digital latch where positive feedback from a cross-coupled pair amplifies a small differential input voltage into full rail-to-rail logic levels.
In this project: Occurs in DLATCH during the hold phase ($CKb = 1$) via cross-coupled transistors M3 and M4, holding the sampled state when M6 conducts.
The fundamental design inequality derived in this project guaranteeing that the small-signal open-loop gain of the cross-coupled regenerative pair exceeds unity ($g_m R_1 > 1$).
In this project: Required $R_1 \ge 7.59\text{ k}\Omega$ for $m=1$, and $R_1 \ge 1.90\text{ k}\Omega$ for $m=4$. Proves mathematically why Kuo et al.'s unit sizing rule fails in 45 nm.
The excess load resistance or loop gain beyond the critical oscillation threshold needed to ensure reliable latching across process, voltage, and temperature variations.
In this project: At multiplier m = 4, theoretical regeneration threshold is R_crit = 1.90 kΩ. Design D uses R = 3.0 kΩ (giving 1.10 kΩ margin, passing 41/45 PVT corners); Design B uses R = 2.0 kΩ (0.10 kΩ margin, passing 21/45 corners).
A formal, verifiable technical statement specifying a functional capability, performance target, or constraint, with an explicit acceptance criterion and evidence file.
In this project: Formulated in v2/REQUIREMENTS.md as 12 formal requirements (SR-01 to SR-12), including SR-01 for divide-by-two functionality, SR-05 for 45-corner PVT sweeps, and SR-08 forbidding post-layout claims without extraction.
A closed loop arrangement of logic elements where the output of each stage feeds the input of the subsequent stage, with an overall feedback loop inversion.
In this project: Four DLATCH stages are connected in a closed ring ($XD1 \to XD2 \to XD3 \to XD4 \to XD1$) with alternating cross-connections, eliminating the need for an 8-latch topology.
A formal project management record cataloging identified technical, security, and schedule risks, assessing likelihood and impact, and defining concrete triggers and mitigations.
In this project: Maintained in v2/RISKS.md across 12 risks (R1 to R12), identifying unmeasured draft claim carryover (R1), missing QRC extraction files (R3), VM lane queue serialization (R4), and repository credential exposure (R9).
Cadence configuration control file for Assura DRC, LVS, and RCX containing execution parameters, cell names, view targets, and rule set inclusions.
In this project: Generated dynamically from template files in `cadence/skill/verify/templates/` (`drc.rsf.in`, `lvs.rsf.in`) by shell wrapper `run_verify.sh`.
A foundry-supplied ASCII rule definition file specifying layer operations, geometric spacing rules, and electrical connectivity extraction statements for Assura.
In this project: GPDK045 rule files include `assuraDRC.rul` (DRC rules), `extract.rul` (device and connectivity extraction), and `compare.rul` (LVS comparison tolerances).
S
The operational phase in a latch where input signals are actively tracked and transferred to intermediate internal nodes under clock control.
In this project: Occurs when $CK = 1$, turning on clock switch M5 and enabling differential pair M1–M2 to track input signals D and Db.
The MOSFET operational state where drain-to-source voltage exceeds saturation voltage ($V_{ds} \ge V_{dsat} = V_{gs} - V_{th}$), causing drain current to saturate and depend primarily on $V_{gs}$.
In this project: Assumed for tail current source M7 in classical text design; detailed operating point analysis in Spectre proved M7 operates in triode, not saturation.
Virtuoso schematic verification utility that detects open pins, floating nets, short circuits, and naming conflicts before netlisting.
In this project: Cleaned to 0 errors and 0 warnings on `DLATCH` and `divider_core` using `clean_floating.il`. Exactly 1 known warning remains on `tb_divider` (instance VDD collides with net VDD).
Spectre simulation executed directly from the Virtuoso schematic database using the Virtuoso netlister and OCEAN rather than a standalone hand-written netlist.
In this project: Simulated `tb_divider` at 1.4 GHz via `run_schematic_tb_param.ocn`, yielding $0.455\text{ mW}$ power and $26.66\text{ ps}$ delay, uncovering the layout-dependent parameter effect.
The autonomous oscillation frequency of a closed-loop divider when clock input amplitude is reduced to zero, caused by positive loop gain.
In this project: In the non-dividing paper sizing case ($R = 1.15\text{ k}\Omega$), the circuit autonomously self-oscillates at 2799.8 MHz ($2.0\times f_{in}$) with $15.3\text{ mV}$ swing.
A transistor and load sizing strategy proposed by Kuo et al. setting unit NMOS width and designing load resistance to equal half of on-resistance ($R_1 = R_{on}/2$).
In this project: Fails in 45 nm GPDK045 due to the 3-device pull-down stack; adapted in this project into the stack regeneration rule ($R_1 \ge 1.1 \times 3R_{on}/m$).
Cadence proprietary Lisp-like extension programming language used to automate and customize all operations across Virtuoso schematic, layout, and database editors.
In this project: Extensively used to automate schematic generation (`build_schematic_v2.il`), wire routing (`wire_schematics.il`), stub cleanup (`clean_floating.il`), and layout placement (`build_layout_v2.il`).
Cadence flagship SPICE-class circuit simulator providing high-accuracy transient, RF, and reliability analysis for advanced nanometer nodes.
In this project: Spectre MMSIM121 (12.1.0) inside CentOS VM performed all simulation characterizations, transient runs, DC operating point sweeps, and 45-corner PVT analyses.
The effective series on-resistance of the three cascaded NMOS transistors in the DLATCH pull-down network (sampling device, clock switch, and tail transistor).
In this project: Calculated as $R_{stack} = 3 R_{on} / m = 6.90\text{ k}\Omega$ for unit multiplier $m = 1$, and $1.725\text{ k}\Omega$ for sized multiplier $m = 4$.
Assura DRC substrate connectivity error occurring when a diffusion layer touches a substrate or well without a valid physical tap contact or when multiple conflicting taps exist.
In this project: Appeared in placement layout DRC: 1 error in Rule 220 (`psubstrate_StampErrorMult`) and 6 errors in Rule 221 (`psubstrate_StampErrorConnect`) due to missing substrate tie routing.
A headless Spectre netlist created for standalone simulation without opening Cadence Virtuoso, using default BSIM4 process parameters.
In this project: Used for all batch parameter sweeps, PVT matrices, and frequency sweeps; produces $0.541\text{ mW}$ power and $25.99\text{ ps}$ delay for Design D.
Mechanical stress exerted on transistor silicon active channels by surrounding Shallow Trench Isolation oxide, altering carrier mobility and threshold voltage based on distance to STI edge.
In this project: Quantified by BSIM4 instance parameters `sa=140n, sb=140n, sd=160n`; identified as one of the primary contributors to the schematic-vs-standalone power difference.
A multi-channel validation algorithm requiring all eight outputs of the divider to toggle at exactly half the input frequency ($f_{in} / 2 \pm 0.1\%$) with swing $\ge 0.1\text{ V}$.
In this project: Implemented in `measure_guest.py` after earlier Q1-only checks yielded false passes on harmonic distortions at $2.0\times f_{in}$.
A multi-agent engineering workflow distributing tasks between specialized LLM agents for VM execution, scripting, text auditing, and verification.
In this project: Haiku workers executed headless VM runs and log shipping; automated quantitative audits (`REVIEW3_numbers.md`, `REVIEW3_prose.md`) were conducted to eliminate promotional bias.
Heavily doped ohmic diffusion regions (p+ in p-substrate, n+ in n-well) that connect bulk silicon to ground (VSS) or power (VDD) to prevent latch-up.
In this project: Missing in early placement PCells, causing Assura DRC substrate stamp errors; required in final physical layout to tie `g45rnsnp` resistor bulk terminals.
The weak diffusion current flowing between drain and source of a MOSFET when gate voltage is below threshold ($V_{gs} < V_{th}$): $I_{sub} \propto \exp[(V_{gs} - V_{th}) / (n k_B T / q)]$.
In this project: Increases significantly at the high temperature corner ($125^\circ\text{C}$), reducing effective logic swing and causing the 4 failing PVT points in Design D.
Power supply voltage variations evaluated during PVT characterization, spanning $\pm 10\%$ around nominal 1.0 V supply.
In this project: Tested at $0.9\text{ V}$ (worst-case drive current), $1.0\text{ V}$ (nominal), and $1.1\text{ V}$ (maximum PDK voltage rating for `nmos1v`).
The positive direct current electrical supply rail providing operating power to the integrated circuit.
In this project: Nominal $V_{DD} = 1.0\text{ V}$ for GPDK045 standard devices, compared to $1.8\text{ V}$ in Kuo et al.'s 180 nm reference paper.
T
A current-regulating element connected between the common-source node of a differential pair and ground, establishing total bias current.
In this project: Implemented using transistor M7 ($W = 480\text{ nm}, L = 90\text{ nm}$) with gate tied to $v_b = 1.0\text{ V}$, operating in triode region.
Operating temperature extremes evaluated during environmental characterization: industrial cold ($-40^\circ\text{C}$), room nominal ($27^\circ\text{C}$), and junction hot ($125^\circ\text{C}$).
In this project: Simulated across all 5 process corners; Design D passed 15/15 corners at $-40^\circ\text{C}$, 15/15 at $27^\circ\text{C}$, and 11/15 at $125^\circ\text{C}$.
The minimum gate-to-source voltage required to invert the silicon surface and create a conducting channel between drain and source.
In this project: Nominal $V_{th} \approx 0.35\text{ V}$ for `g45n1svt` devices at $27^\circ\text{C}$; drops at high temperature ($125^\circ\text{C}$) and varies across process corners.
The property that enables any reported number, plot, or conclusion to be tracked directly to an underlying raw simulation log, netlist, and rerun script.
In this project: Every number in the explainer and report maps directly to a specific log or JSON artifact in `results/gpdk045/` or `report/gpdk045/report_data.json` with an exact command to reproduce it.
A structured comparative evaluation of alternative design choices against weighted technical feasibility, cost, headroom, and risk criteria.
In this project: Conducted in v2/TRADE_STUDY.md across 7 architecture candidates (E1 to E7); E1 (sizing exploration) scored 50/60 as the recommended primary path, whereas cascode tail (E2) scored 18/60 and was rejected.
The MOSFET operational state where drain-to-source voltage is smaller than saturation voltage ($V_{ds} < V_{dsat} = V_{gs} - V_{th}$), acting as a voltage-controlled linear resistor.
In this project: Tail transistor M7 operates deeply in triode ($V_{ds} = 8\text{--}67\text{ mV} \ll V_{dsat} = 358\text{ mV}$), acting as an ohmic degeneration resistor rather than an ideal current source.
The continuous frequency span over which a circuit maintains correct operation, expressed in MHz/GHz or as percentage fractional bandwidth: 2 * (f_max - f_min) / (f_max + f_min) * 100%.
In this project: Kuo et al. reported a 173.3% range (100 MHz to 1.4 GHz) in 180 nm; in 45 nm, our sweeps demonstrate that frequency range depends on sizing (Design D operates up to 5.5 GHz; Design B up to 6.0 GHz).
U
The working unit-device frequency divider circuit using $m = 1$ transistors and $8\text{ k}\Omega$ load resistors operating at 1.4 GHz.
In this project: Yields 700.0 MHz output with $0.1375\text{ mW}$ power, $62.21\text{ ps}$ delay, $8.557\text{ fJ}$ PDP, $0.2289\text{ V}$ swing, and $26.42^\circ$ phase error, serving as the benchmark without sizing.
V
The drain-to-source voltage at which electron velocity saturation occurs in short-channel MOSFETs, demarcating the boundary between triode and saturation regions.
In this project: Extracted as $V_{dsat} = 0.3578\text{ V}$ ($357.8\text{ mV}$) for M7 in Spectre; transient tail node voltages stay below $68.2\text{ mV}$, proving M7 remains in triode.
A tabular cross-reference mapping each system requirement to its acceptance criterion, verification method (test, inspection, analysis), and specific evidence log path.
In this project: Documented in v2/VERIFICATION_MATRIX.md to verify that all system requirements (SR-01 to SR-12) are satisfied before manuscript submission.
A conductive vertical metal plug etched through inter-layer dielectric oxide connecting two adjacent horizontal metal routing layers.
In this project: Via1 connects Metal1 to Metal2; the automated router inserted 37 vias during the `DLATCH/layout_routed` routing pass.
Type-2 hardware virtualization hypervisor hosted on the Linux workstation to run the legacy 32-bit CentOS 6.5 Cadence environment.
In this project: Managed headless via `VBoxManage controlvm CadenceVM` to dispatch keyboard strings and execute automated SKILL and Spectre batch runs.
Cadence graphical waveform display and signal analysis environment embedded in Virtuoso for viewing transient, frequency, and eye diagram waveforms.
In this project: Used to capture full-screen multi-phase transient waveforms (`viva_tran_D_1p4GHz_full.png`) displaying all 8 divided outputs and differential clock inputs.
Assura hierarchical compiled database format that stores schematic netlist structures for direct comparison against extracted layouts during LVS.
In this project: Produced by running `dfIIToVldb` on `divider_lib/DLATCH` schematic views inside the VM prior to executing `avLVS()`.
The allowable DC voltage range across active transistors in a series stack before devices are forced into linear/triode operation or encounter supply rail limits.
In this project: Under 1.0 V supply with a three-transistor series stack and load resistor IR drop, available headroom above the tail node is only 8 to 68 mV, precluding cascode topologies that require ~700 mV headroom.
W
Threshold voltage and mobility shifts in MOSFETs caused by scattered high-energy dopant ion scattering off photoresist edges during retrograde well implantation.
In this project: Quantified by BSIM4 instance parameters `sca=226.00151, scb=0.11734, scc=0.02767` extracted by Virtuoso, explaining lower drive current and power in schematic runs.